Openlayer MCP

Model Context Protocol server for the Openlayer platform. It lets MCP clients — Cursor, VS Code, Claude Desktop, and other AI assistants — work with Openlayer in natural language.

With it, an assistant can:

• Projects & versions — list and create projects; push and inspect commits.
• Data sources — list, create, and inspect data sources (inference pipelines), and fetch their production traces, sessions, and per-user activity, with filters; comment on traces as the connected user.
• Tests — list and create tests across integrity, consistency, and performance (including LLM-as-a-judge), trigger evaluation, read results, and drill into the rows that failed.
• Anything else — call any Openlayer API endpoint directly, and search the Openlayer documentation.

Tools are annotated as read-only or destructive, so clients can auto-approve reads and confirm writes.

Tool names and schemas may evolve; clients should discover the available tools dynamically rather than hard-coding them.

📖 Full setup, the complete tool reference, and examples: openlayer.com/docs/openlayer-mcp

Installation

uv add openlayer-mcp     # or: pip install openlayer-mcp

Configuration

Register the server with your client and set your API key.

Claude Code (CLI):

claude mcp add openlayer --env OPENLAYER_API_KEY=YOUR_OPENLAYER_API_KEY -- uvx openlayer-mcp

Cursor, Claude Desktop, or VS Code (JSON config):

{
  "mcpServers": {
    "openlayer": {
      "command": "uvx",
      "args": ["openlayer-mcp"],
      "env": {
        "OPENLAYER_API_KEY": "YOUR_OPENLAYER_API_KEY"
      }
    }
  }
}

• OPENLAYER_API_KEY (required) — create one in your Openlayer workspace settings.
• OPENLAYER_BASE_URL (optional) — point at a self-hosted Openlayer instance; defaults to Openlayer Cloud.

Verify it's working by asking your assistant: "List my Openlayer projects."

You can also run it as a standalone server:

uv run -m openlayer_mcp

Transport

The server runs over stdio by default (for local IDE/desktop clients). For hosted or multi-client deployments, set OPENLAYER_MCP_TRANSPORT to streamable-http, http, or sse:

OPENLAYER_MCP_TRANSPORT=streamable-http uv run -m openlayer_mcp

Remote hosting

This repo also has a separate remote server package (server/, not published to PyPI) that runs the same tools as an OAuth-protected Streamable HTTP connector for Claude. Its AWS infrastructure and deploy live in the in-repo CDK app under awscdk/; its Docker image is published to the private Docker Hub repo openlayer/openlayer-mcp for self-hosting/on-prem customers we grant pull access to. See server/README.md.

Documentation

For the full guide and the complete list of available tools, see openlayer.com/docs/openlayer-mcp.